lncRedibly versatile: biochemical and biological functions of long noncoding RNAs

Abstract

Long noncoding RNAs (lncRNAs) are transcripts that do not code for proteins, but nevertheless exert regulatory effects on various biochemical pathways, in part via interactions with proteins, DNA, and other RNAs. LncRNAs are thought to regulate transcription and other biological processes by acting, for example, as guides that target proteins to chromatin, scaffolds that facilitate protein–protein interactions and complex formation, and orchestrators of phase-separated compartments. The study of lncRNAs has reached an exciting time, as recent advances in experimental and computational methods allow for genome-wide interrogation of biochemical and biological mechanisms of these enigmatic transcripts. A better appreciation for the biochemical versatility of lncRNAs has allowed us to begin closing gaps in our knowledge of how they act in diverse cellular and organismal contexts, including development and disease.

INTRODUCTION

DNA, RNA, and proteins play critical roles in living cells, but RNA is uniquely able to both carry genetic information in the form of nucleotide sequence and also perform biochemical functions due to its ability to fold into complex tertiary structures. The long-held view that biochemical pathways are the domain of proteins has been reconsidered in recent years, as more roles have been attributed to once-thought non-functional RNA molecules that are not translated into proteins [1]. This paradigm shift was in part spurred by advances in large-scale analyses of gene expression, first provided by tiling oligonucleotide microarrays followed by even more powerful next generation sequencing technologies, which revealed active transcription in ever-growing swaths of noncoding regions of the genome [2–6]. Only a very small portion of this widespread noncoding transcription could be assigned to conserved RNAs with well-known functions, such as ribosomal RNAs, transfer RNAs, small nuclear and nucleolar RNAs, and microRNAs. Most of the remaining noncoding transcripts were grouped in a category comprising RNAs longer than 200 nucleotides, puzzlingly lacking an obvious function, that were originally termed simply “large” or “long” noncoding RNAs (lncRNAs) [7–9].

Thousands of lncRNAs have been annotated in animals, including various mammals [6, 9–12], zebrafish [13], and insects [14–16]. Although plant and yeast genomes also contain lncRNAs, this review focuses on studies in humans, mice, and fruit flies, with only passing mentions of lncRNAs from plants and fungi, which have been reviewed elsewhere [17–19].

Some lncRNAs had previously been identified by conventional methods. Arguably, the first molecule in this category was found in 1990, when the imprinted H19 transcript was noted to lack a long open reading frame and association with the translational machinery, suggesting that the RNA itself was functional, not an encoded protein product [20, 21]. Similarly, the lncRNA XIST, which is critical for X chromosome inactivation [22, 23], was discovered in 1991 and reported to lack coding potential [24, 25]. However, the functional classification of most lncRNAs has been hindered thus far by low expression levels, a lack of sequence conservation, and incomplete annotation in many organisms. In recent years, improved biochemical, genetic, and computational methods have given researchers more tools to study lncRNAs and their regulatory roles, to dissect their biochemical mechanisms of action, and to understand their contributions to cancer [26, 27] and other diseases [28, 29].

Many reviews have been written on various aspects of lncRNA biology [30–39]. In this one, we focus on relatively recent new insights into the function of nuclear lncRNAs in animals, with a particular focus on mechanistic studies and newly developed biochemical methods.

Biochemical mechanisms of lncRNA function

The pioneering papers reporting the existence of lncRNA on a broad scale, as well as the results of the first perturbation studies, suggested that this broad category of transcripts might share a common biological function in gene regulation [9, 40–42], as do known classes of small noncoding RNAs, such as microRNAs, and piwi-associated RNAs [43]. However, it has become clear that the only thing that all lncRNAs have in common is their lack of coding potential. Many different subtypes of lncRNAs exist, such as promoter-associated, enhancer-derived, antisense, and interleaving [30], each likely carrying out diverse functions in different molecular contexts. Even broader diversity exists among the interleaving lncRNA subtype (lincRNAs), which might possess a variety of biochemical and cellular functions and will likely require individual characterization. Nonetheless, several central themes are emerging for potential broad molecular roles that lncRNAs might be playing in the cell. In this review, we mainly elaborate on lncRNAs that have been proposed to regulate chromatin structure and transcription; however, roles for lncRNA in a diverse set of crucial processes such as alternative splicing [44–46], translation [47], and genome stability [48, 49] have been identified, and more surely remain to be discovered. Below we review new evidence in support of two previously proposed [50] biochemical mechanisms of lncRNA function, chromatin recruitment and protein complex scaffolding, as well as their emerging role as potential agents for nuclear organization and phase separation. We emphasize that these mechanisms of action need not be mutually exclusive, as the same lncRNA might have different biochemical functions (Figure 1).

Figure 1. Biochemical functions of lncRNAs.

(A) Some lncRNAs form RNA/DNA triplexes (top), providing a potential mechanism for specific targeting of proteins to chromatin. LncRNAs discussed in the text are highlighted along with their associated proteins and target genes. LncRNAs could also recognize target regions on chromatin via Watson-Crick interactions with the DNA, forming R-loops (middle). TERRA forms an R-loop at its own locus at the telomeric region of chromosomes, activating the DNA damage response. Some lncRNAs are thought to guide chromatin-modifying complexes and transcription factors to target genes (bottom).

(B) A number of lncRNAs act as scaffolds to facilitate interactions between proteins, as in the NORAD-TOP1-RBMX complex that promotes genome stability and the rOX2-MSL-MLE complex involved in dosage compensation in Drosophila. Some lncRNAs have multiple biochemical functions, as with the scaffolding lncRNA GUARDIN, which also promotes genome stability through sequestering miR-23a.

(C) Some lncRNAs may affect gene expression by regulating genome. HIDALGO, ThymoD and HOTTIP regulate looping interactions at their own loci, while EVF2 and FIRRE are organizers of genome architecture on a larger scale. Some of these lncRNAs are aided by architectural proteins, such as CTCF and cohesin.

(D) MALAT1 and NEAT1 lncRNAs are important for assembly of two phase-separated bodies in the nucleus, speckles and paraspeckles, respectively. Speckles contain proteins involved in transcription and splicing; MALAT1 regulates the localization and distribution of these proteins. NEAT1 acts as a scaffold to bring together members of the RNA production and processing machinery in paraspeckles.

Chromatin guides

Among the first molecular function assigned to lncRNAs was acting as a guide to recruit protein complexes, in particular those that participate in gene regulation, to their target sites on chromatin [50, 51] (Figure 1A). Various models have been proposed for how lncRNAs localize to specific sites on chromatin to facilitate protein recruitment, either proximal to their transcribed locus or at distal loci. LncRNA-mediated targeting of proteins to chromatin is composed of two distinct processes: the interaction of a lncRNA with the DNA sequence on one side and with protein effectors on the other.

Finding targets on DNA

LncRNAs can target specific sequences on chromatin either via direct association with DNA or indirectly as part of a protein–RNA complex. As previously hypothesized [52, 53], sequence-driven recognition of genomic target sites by lncRNAs could occur either at intact double stranded DNA (triplex mode) or by displacing a DNA strand (R-loop mode). DNA–RNA triplexes (Figure 1A) are well-characterized structures formed by the sequence-specific interaction of an RNA strand with the major groove of a double-stranded DNA helix [54, 55]. These structures are stabilized by noncanonical Hoogsteen hydrogen bonding between an RNA base and the canonical Watson-Crick base pair of DNA. Triplexes can be either parallel or antiparallel, based on the orientation of the RNA strand relative to the target DNA strand [54]. One example of a lncRNA binding to chromatin via a triplex interaction occurs at the human DHFR locus, where an upstream ncRNA forms an RNA–DNA triplex at the promoter, preventing transcription by disrupting the pre-initiation complex [56]. Another example is found at the rDNA locus in mouse, from which ribosomal RNAs are transcribed. Here, lncRNAs from the sense (pRNA) and antisense (PAPAS) strand were shown to form triplexes locally at its promoter and to recruit to the locus the DNA methyltransferase DNMT3B and the chromatin remodeling complex NuRD, respectively [57, 58], with the result of repressing rDNA transcription. Additional lncRNAs reported to form triplexes at distal loci include the developmental regulator FENDRR, the imprinted lncRNA MEG3, and the irradiation-responsive PARTICLE [59–62], all of which were also reported to recruit the Polycomb repressive complex PRC2 (but see below for a discussion on the complexity of PRC2–RNA interactions). Despite these intriguing examples, clear rules have yet to emerge that would allow prediction of lncRNAs binding to genomic targets at distal loci based on the formation of DNA–RNA triplexes.

An alternative means by which lncRNAs could be recruited to chromatin is by forming DNA–RNA hybrid structures called R-loops (Figure 1A), whereby a complementary RNA directly pairs to DNA by displacing one of its strands [63]. This process normally occurs co-transcriptionally, when a nascent transcript binds to its template DNA. A well-characterized example is the Terra family of lncRNAs in S. cerevisiae, which regulate telomere length and cell senescence via R-loop formation [64]. Other studies in yeast have reported that lncRNA-mediated R-loops can occur post-transcriptionally at distal loci, but whether this is the case in higher organisms remains unknown [65, 66]. Emerging new technologies to map R-loops with high resolution and sensitivity in mammalian cells [67–70] might help answering this important question.

Some lncRNAs utilize spatial proximity to localize to specific chromatin sites. For example, local chromatin topology directs the spreading of XIST on the X chromosome and insertion of an Xist transgene at a different genomic locus changes its early localization pattern [71]. Whether this proximity model can be extended to other lncRNAs is not known.

Chromatin proteins that bind to lncRNAs

Typically, lncRNA are more dynamically expressed during development, and maintain more spatial specificity than protein-coding genes [10, 11]. Together with their ability to guide proteins to chromatin, this makes lncRNA prime candidates for factors that contribute target specificity to epigenetic regulators, which often lack DNA-binding domain, during development and in different cell types. A role for lncRNAs in recruiting the epigenetic silencer PRC2 to chromatin has been hypothesized for many years, though questions remain about the specificity of lncRNA binding and mechanisms of recruitment [39, 72, 73]. One of the first lncRNAs classified as functional was HOTAIR, which was shown to silence HOXD10 by facilitating the recruitment of the epigenetic silencing complex PRC2 through direct interactions [51]. Although following studies cast some doubt on the biochemical details of the initial model [74] and have not come to an agreement on the phenotypic consequences of the genetic deletion of Hotair in mice [75–78], the pioneering work on HOTAIR spurred a search for other lncRNA–protein interactions that might underpin the recruitment of transcriptional regulators to chromatin, which in many cases pointed back to PRC2 and its affinity for various lncRNAs.

In fact, several subunits of PRC2 as well as PRC1 (another well-studied Polycomb repressive complex [79]) have been reported to bind to RNA in mammals, including EZH2, JARID2, SUZ12, CBX7, and SCML2 [80–87]. Similarly, the plant lncRNAs COLDAIR, and COLDWRAP, which are involved in seasonal regulation of flowering in Arabidopsis thaliana, were observed to interact with PRC2 [88, 89]. In animals, members of the trithorax group of proteins, which counter Polycomb-mediated epigenetic silencing [90] can also bind to lncRNAs. HOTTIP is a lncRNA transcribed from the Hoxa locus that binds to WDR5, a subunit of the H3K4 methyltransferase MLL complex [91, 92]. Other examples of lncRNAs that guide deposition of epigenetic histone marks are AIR, which binds to the G9a H3K9 methyltransferase [93], and PAUPAR, which affects levels of the repressive H3K9me3 mark at target promoters by interacting with KAP1/TRIM28 [94, 95].

In addition to regulating the deposition of histone modifications, lncRNAs can affect another key epigenetic modification, DNA methylation, via the recruitment of DNA methyltransferases (DNMTs). In addition to the pRNA from the rDNA locus mentioned above [57], the lncRNAs ecCEBPA and DALI repress methylation at neighboring gene by interacting with and inhibiting DNMT1 [96, 97]. Furthermore, many lncRNAs involved in imprinting regulate DNA methylation at their associated protein-coding genes and might do so by directly or indirectly recruiting DNMTs [98].

In addition to guiding chromatin-modifying complexes to their targets, lncRNAs can participate in the recruitment of transcription factors, although these proteins also have a built-in ability to recognize DNA motifs embedded in the genomic sequence. The intronic lncRNA PANCT1 was originally identified in a screen for positive regulators of the pluripotency factor Pou5f1/Oct4 [99]. PANCT1 binds to the protein product that originates from its own locus, TOBF1, and increases its occupancy at target genes on chromatin. Interestingly, promoters for these genes are enriched for the Oct-Sox sequence motif, which is also embedded in the lncRNA itself. Rescue of TOBF1 binding can be achieved with overexpression of wild type PANCT1, but not a mutant with a scrambled Oct-Sox motif [100], demonstrating that the lncRNA can function at distal loci and suggesting that PANCT1 recognizes target sequence on DNA directly, perhaps by forming an R-loop (see above). The lncRNA RMST is another effector of cell fate choice that facilitates binding of SOX2 to the promoters of neurogenic transcription factors via direct association [101]. Finally, some lncRNAs prevent the recruitment of transcription factors to chromatin, as in the example of the lncRNA PANDA, which inhibits chromatin binding of the transcription factor NF-YA [102].

RNA binding is a property shared by many chromatin-modifying complexes and chromatin-associated proteins [40, 41, 103, 104], suggesting that the list of protein–RNA interactions with an epigenetic role is destined to grow. However, years of research on RNA-mediated regulation of chromatin function have taught us that isolated reports of lncRNA–protein interactions correlated with transcriptional effects might not be sufficient to draw strong conclusions on the biochemical and biological function of a lncRNA. Fortunately, emerging technologies will provide us with the ability to detect and manipulate lncRNA–protein interactions with ever increasing resolution and accuracy (see below) and will enable scientists in the field to perform rigorous rescue and tethering experiments using mutant proteins defective in RNA binding as controls. These experiments will allow us to confirm or refute proposed models for the function of lncRNAs within chromatin-associated complexes.

Scaffolds for complexes

A diverse collection of lncRNAs scaffold interactions between proteins and protein complexes (Figure 1B), affecting epigenetic regulation of gene expression, genome stability, nuclear organization, and microRNA biogenesis.

Drosophila achieves dosage compensation of the sex chromosome by doubling the transcriptional output from the single male X chromosome. This function is carried out by the dosage compensation complex (DCC), which is formed by the histone acetyltransferase Males-absent on first (MOF), four different Male sex lethal proteins (MSL1–4) and the RNA helicase Maleless (MLE) [105]. In addition to these protein components, spreading of DCC to the entire X chromosome requires the presence of at least one of two lncRNAs, roX1 and roX2 [106, 107]. Protein-RNA crosslinking revealed that two stem loops on roX2 act as a switch for the binding of MSL2: when MLE is bound to roX2, one of the stem-loops is unwound, allowing for the formation of the other stem-loop and promoting binding of MSL2 [108]. The presence of RNA, is required to stabilize the association of MLE with the other DCC proteins [109], suggesting the possibility that roX2 (and perhaps roX1) provides a scaffolding functions within this essential complex.

Another epigenetic regulator, PRC2, whose complex relationship with RNA was briefly discussed above, has also been shown to interact with scaffolding lncRNAs. HOTAIR was shown to mediate interactions between PRC2 and the histone demethylase LSD1 [110], whereas the imprinted lncRNA MEG3 was shown to facilitate interactions of the core PRC2 subunit EZH2 with the accessory subunit JARID2 [83].

The lncRNA NORAD was first shown to interact with PUMILIO proteins and proposed to sequester them to promote the stability and translation of PUMILIO mRNAs targets [111]. However, casting doubt on these initial conclusions, RAP-MS (described later) and biochemical studies revealed that NORAD is required for the assembly of a novel complex, termed NARC1, which includes RBMX, TOP1, ALYREF and the PRPF19–CDC5L complex, all of which had previously known roles in suppressing genomic instability [49]. The formation of this complex requires NORAD, which directly binds to RBMX. NORAD depletion causes replication-fork velocity reduction and cell-cycle defects, phenotypes similar to deletion of TOP1 or RBMX, two proteins in the NORAD-dependent complex, thus indicating a role in genome stability for the NORAD transcript. The Norad knockout phenotype of segregation defects was rescued by the reintroduction of the full-length transcript, confirming its ability to act at the RNA level.

Other scaffolding lncRNAs include NEAT1, which organizes paraspeckles, a nuclear subdomain involved in splicing regulation [112] (see below) and regulates the interaction between paraspeckle proteins and the Microprocessor complex, thus enhancing pri-microRNA processing [113], and the lncRNA GUARDIN, which promotes heterodimerization between BRCA1 and BARD1. This interaction stabilizes BRCA1 by preventing its polyubiquitination [114]. GUARDIN can also sequester miR-23a, stabilizing the shelterin complex member TRF2 [114], demonstrating the ability of a single lncRNA to act via multiple non-exclusive biochemical mechanisms.

Overall, experimental evidence is growing for a pervasive scaffolding function of lncRNAs, which was previously postulated [50, 115]. In addition to the candidate-based studies reviewed above, a recent proteome-wide analysis revealed that ~20% of cellular protein complexes are sensitive to RNase A treatment [116], suggesting that RNA is a component of many mature complexes or at a minimum participates in their assembly. What portion of these scaffolding RNAs are lncRNAs remains to be determined.

Architects of the genome

Another increasingly appreciated function of lncRNAs is the regulation of chromosome conformation [32] (Figure 1C). Xist-mediated repression is associated with a general reshaping of the X chromosome in space [117, 118] and with its relocalization to the nuclear lamina, possibly via a direct interaction of Xist with the lamin B receptor (LBR) [119–121]. Regulating enhancer-promoter contacts also provides a mechanism for controlling gene expression locally. HOTTIP was implicated as a factor in chromatin looping at the Hoxa locus [91], and two “activatory” lncRNAs, ncRNA-a3 and ncRNA-a7 facilitate the formation of chromatin loops between their own locus and target genes, dependent on their interaction with the Mediator complex [122].

A class of lncRNAs that may share this function is chromatin-enriched lncRNAs, or cheRNAs [123, 124]. Many cheRNAs overlap enhancers but have several properties that distinguishes them from canonical enhancer-associated RNAs [125], including polyadenylation, H3K4me3-marked promoters, and a specific strand bias. HIDALGO is a cheRNA derived from read-through transcription of the hemoglobin gene Hgb1, and is required for the interaction between Hgb1 and its enhancer during K562 differentiation. Silencing of Hidalgo using CRISPR interference (CRISPRi) attenuated the contacts between Hgb1 and its two interacting partners, which include the Hidalgo promoter, and reduced Hgb1 expression. Antisense oligonucleotide-mediated depletion of HIDALGO recapitulated this effect, demonstrating a role for the RNA.

The phenomenon of lncRNAs activating genes near the site of their transcription is widespread [126], perhaps suggesting that the facilitation of looping either by the act of transcription or by the lncRNA is responsible for local regulation of gene expression. Indeed, lncRNAs that act locally are enriched at the boundaries of topological associated domains [127, 128], suggesting a role for lncRNAs in regulating local genome architecture.

One way lncRNAs might shape chromatin conformation is by interacting with known architectural proteins, such as cohesin and CTCF [129, 130]. The lncRNA ThymoD activates expression of the T-cell specific gene Bcl11b by facilitating a promoter-enhancer loop [131]. ThymoD is transcribed from a downstream control region of Bcl11b and is required for interaction between Bcl11b and its enhancer. Attenuation of ThymoD transcription led to reduced CTCF and cohesin occupancy at the Bcl11b control region, abolishing the interaction and reducing Bcl11b expression. Another example is EVF2, transcribed from an ultraconserved super-enhancer, which maintains the gene expression program in mouse interneurons [132]. EVF2 localizes with cohesin subunits at targets genes, regulating their interactions of the super-enhancer from which it is transcribed.

The FIRRE lncRNA has been proposed to function as an organizer of interchromosomal interactions as well as a facilitator of correct nuclear localization of the inactive X chromosome. FIRRE localizes to the X chromosome from which it is transcribed, but also to distal autosomes, bringing these regions into nuclear proximity with each other [133]. FIRRE binds both cohesin and CTCF on the inactive X, contributing to its anchoring to the nucleolus [134]. Knockdown of FIRRE caused deregulation of both X-linked and autosomal genes, perhaps due to the loss of H3K27me3 on the inactive X chromosome. Interestingly, although CTCF occupancy and long scale interactions on the X chromosome are disrupted by Firre knockout, the overall structure of topologically associated domains (TADs) is unchanged [135].

The RNA binding activity of CTCF was initially characterized in the context of its interaction with the antisense lncRNA WRAP53 [136], and it was mapped to an RNA-binding region that was recently shown to be required for a large fraction of CTCF-dependent loops, suggesting the possibility that many more lncRNAs might control genome organization by interacting via CTCF [137, 138].

Organizers of phase separation

Phase separation allows for functional compartmentalization in the cell, resulting in droplets where key factors are concentrated, thereby facilitating biochemical processes. Proteins with intrinsically disordered domains, or low complexity domains, interact to form “hubs,” which are ensembles of phase-separated molecules with hydrogel-like properties [139]. The assembly of stress granules, a phase-separated membrane-less organelle, is facilitated by the presence of RNA [140]. Many proteins with disordered regions interact with RNA [104, 141, 142], suggesting that the diversity of lncRNA sequences, expression patterns, and protein-binding properties might contribute to specifying compositionally and functionally distinct phase-separated compartments.

Paraspeckles and speckles are two such compartments that concentrate proteins in the nucleus in phase separated domains [143, 144] (Figure 1D). The lncRNA MALAT1 (also known as NEAT2) has long been known to associate with speckles in the nucleus [145], which contain members of transcription and splicing machinery [146]. MALAT1 regulates the localization and distribution of active splicing factors in speckles, thereby affecting the alternative splicing of pre-mRNAs and its acute depletion by knockdown is lethal in transformed HeLa cells [147], but a complete deletion is surprisingly compatible with mouse development [148], and therefore key questions about its function in vivo. In contrast, NEAT1 is known to be required for the formation of paraspeckles [143] which are almost completely abolished in NEAT1 knockdown cells [149]. Paraspeckles contain members of the DBHS protein family, which play a part in various aspects of RNA production and processing [150]. Both MALAT1 and NEAT1 were reported to localize to chromatin at active sites of transcription [151] and it is tempting to speculate that these lncRNAs might facilitate phase transition processes, which are increasingly appreciated to play a role in transcriptional regulation [152].

The functional domains of NEAT1 have recently been explored in great depth using systematic deletions of different regions of the transcript [112]. NEAT1 has two isoforms; the longer isoform is stabilized by a triple helix, and is required for paraspeckle formation. The middle domain of NEAT1 contains long repetitive sequences, and is necessary and sufficient (along with the terminal 5’ end and the stabilizing triple helix) for the formation of ordered paraspeckles. Treatment with the known disruptor of phase-separated structures 1,6-hexanediol dissolved intact paraspeckles and abolished the co-localization of NEAT1 with the essential paraspeckle protein NONO. In addition, cells expressing a deletion mutant of NEAT1 that lacks protein binding domains do not form normal paraspeckles. Tethering paraspeckle proteins to NEAT1 rescues normal paraspeckle formation; however, the NOPS dimer formation domain of NONO is required to successfully rescue the phenotype. These observations suggest that NEAT1 recruits NONO, which then initiates formation of the paraspeckles protein complexes, using NEAT1 as a scaffold.

Finally, it has been recently hypothesized that XIST might silence the X chromosomes by forming phase-separated “silence granules”, due to its interactions with disordered proteins, including FUS and hnRNPA2 [22, 153]. Although experimental evidence is required to support this suggestion, it provides an interesting new perspective on the biochemistry of XIST-mediated silencing.

“The company you keep” – techniques to study the lncRNA interactome

The common theme in the biochemical processes discussed above is that the molecular functions of lncRNAs emerge from their interactions with other proteins and DNA. Techniques to study protein–protein and DNA–protein interactions in both candidate-based and unbiased fashions have been available for decades, and have been coupled with high-throughput approaches in genomics and proteomics in recent years. Although similar approaches for lncRNA discovery have been lagged behind, the last few years have seen great advances in techniques to map protein–RNA and DNA–RNA interactions. The definitions for the many acronyms of methods mentioned in the following sections are collected in Table 1 along with the relevant citations.

Table 1.

Biochemical methods to identify lncRNA interactions.

Method	Full name	Use	Citation
DNA–RNA
CHAR-seq	Chromatin-Associated RNA sequencing	genome-wide DNA–RNA contacts	Bell et al., Elife 2018
CHART	Capture Hybridization Analysis of RNA Targets	candidate RNA localization on chromatin	Simon et al., PNAS 2011
ChIRP	Chromatin Isolation by RNA Purification	candidate RNA localization on chromatin	Chu et al., Mol Cell 2011
dChIRP	domain-specific Chromatin Isolation by RNA Purification	candidate RNA localization on chromatin	Quinn et al., Nat Biotechnol 2014
GRID-seq	Global RNA Interactions with DNA by deep sequencing	genome-wide DNA–RNA contacts	Li et al., Nat Biotechnol 2017
MARGI	Mapping RNA-genome Interactions	genome-wide DNA–RNA contacts	Sridhar et al., Curr Biol 2017
RAP	RNA Antisense Purification	candidate RNA localization on chromatin	Engreitz et al., Science 2013
RNA-DamID	RNA-DNA adenine methylase identification	candidate RNA localization on chromatin	Cheetham et al., Nat Struct Mol Biol 2018
SPRITE	Split Pool Recognition of Interactions by Tag Extension	genome-wide DNA–RNA contacts	Quinodoz et al., Cell 2018
protein–RNA
IPL	in vivo Proximity Labeling	proximity labeling for protein–RNA	Beck et al., J Proteome Res 2014
CARIC	click Chemistry-assisted RNA Interactome Capture	proteome-scale identification of all RBPs	Huang et al., PNAS 2018
CLIP	ultraviolet Crosslinking and Immunoprecipitation	discovery of RNA interactors of a candidate protein	Ule et al., Science 2003
eCLIP	enhanced CLIP	discovery of RNA interactors of a candidate protein	Van Nostrand et al., Nat Methods 2016
GoldCLIP	Gel-omitted ligation-dependent CLIP	discovery of RNA interactors of a candidate protein	Gu et al., Genom Prot Bioinf 2018
HITS-CLIP	High-Throughput Sequencing of RNA isolated by CLIP	discovery of RNA interactors of a candidate protein	Licatalosi et al., Nature 2008
iCLIP	individual nucleotide resolution CLIP	discovery of RNA interactors of a candidate protein	König et al., Nat Struct Mol Biol 2010
iDRIP	identification of Direct RNA Interacting Proteins)	discovery of protein interactors of a candidate RNA	Minajigi et al., Science 2015
RaPID	RNA-protein Interaction Detection	discovery of protein interactors of a candidate RNA	Ramanathan et al., Nat Methods 2018
RBDmap	RNA-binding Domain Mapping on a proteome scale	proteome-scale identification of polyA-bound RBPs and domains	Castello et al., Mol Cell 2016
RBR-ID	Proteomic Identification of RNA-binding Regions	proteome scale identification of RBPs and domains	He et al., Mol Cell 2016
RICK	newly transcribed RNA Interactome capture with Click chemistry	proteome scale identification of nascent RNA-associated RBPs	Bao et al., Nat Methods 2018
RIP-seq	RNA Immunoprecipitation and sequencing	discovery of RNA interactors of a candidate protein	Zhao et al., Mol Cell 2010
uvCLAP	Ultraviolet Crosslinking and Affinity Purification	discovery of RNA interactors of a candidate protein	Maticzka et al., Nat Commun 2018
XRNAX	protein-crosslinked RNA extraction	proteome scale identification of RBPs and domains	Trendel et al., Cell 2018
RNA–RNA
CLASH	Crosslinking Ligation and Sequencing of Hybrids	candidate protein-associated RNA-RNA interactions	Helwak et al., Cell 2013
COMRADES	Crosslinking of Matched RNAs and Deep Sequencing	identification of RNA duplex interactions of candidate RNAs	Ziv et al., Nat Methods 2018
LIGR-seq	LIGation of interacting RNA and high-throughput sequencing	trasncriptome-wide identification of RNA duplexes	Sharma et al., Mol Cell 2016
PARIS	Psoralen Analysis of RNA Interactions and Structures	transcriptome-wide identification of RNA duplexes	Lu et al., Cell 2016
RAP-RNA	RAP for RNA-RNA interactions of a target RNA	intermolecular RNA interactions of candidate RNA	Engreitz et al., Cell 2013

Mapping lncRNAs on chromatin

The biochemical functions of lncRNAs go hand in hand with their subcellular localization. While some lncRNAs are exported to the cytoplasm, many remain in the nucleus, where they can interact with DNA and protein complexes. LncRNAs are generally enriched in the nucleus compared to protein-coding RNAs [6, 154], and recent efforts fusing typically cytoplasmic transcripts to different regions of known nuclear lncRNAs have identified sequence motifs responsible for lncRNA nuclear retention [155, 156]. These motifs demonstrated higher conservation than other lncRNA regions, suggestive of functionality [156]. One such motif found was an Alu-derived sequence, called SIRLOIN for SINE-derived nuclear RNA localization [155]. The nuclear retention of SIRLOIN-containing RNAs was mediated by the RNA-binding protein HNRNPK.

The longstanding hypothesis that lncRNAs participate in gene regulation [30, 157, 158] is supported by the observation that many of them are enriched in the chromatin fraction [123, 124]. Mapping lncRNAs along the genome is thus critical for understanding this suspected function. Various methods have been developed toward this end, including CHART, ChIRP, and RAP [71, 159, 160]. These methods identify genome-wide lncRNA binding sites by chemical crosslinking followed by the capture of target RNAs using antisense biotinylated probes (Figure 2A). DNA that co-purifies with the lncRNA of choice is sequenced and regions of enrichment are identified, similar to existing strategies to localize protein on chromatin, such as chromatin immunoprecipitation. While conceptually similar, these methods differ in probe design, crosslinking strategies, and DNA elution conditions. RAP uses long 120 nt probes that tile the entire RNA sequence, whereas CHART and ChIRP opt for shorter 20–25 nt probes. ChIRP employs glutaraldehyde crosslinking to fix DNA–RNA contacts, which can potentially capture interactions over long distance, in contrast to CHART’s usage of formaldehyde, which crosslinks across shorter distances. RAP uses both formaldehyde and disuccinimidyl glutarate to capture both shorter- and longer-range contacts [161]. dChIRP is a further refinement of the ChIRP method, which probes individual lncRNA domains, providing a higher resolution map of interactions [162].

Figure 2. Methods for mapping the lncRNA interactome.

(A) Outline of methods to study lncRNA localization on chromatin. RAP, ChIRP, and CHART (left) use chemical crosslinking and biotinylated probes to capture a candidate lncRNA crosslinked to chromatin and identify its genomic targets by sequencing the associated DNA. In RNA DamID (middle) cells express a lncRNA of interest fused to MS2 stem-loop sequences and a E. coli adenine methyltransferase (Dam) fused to the MS2 coat protein. This results in methylation at lncRNA-bound genomic sites in vivo. Methylated DNA is then isolated with a methylation-sensitive restriction enzyme and sequenced. MARGI, GRIDseq, and CHARseq (right) use a biotinylated oligonucleotide bridge to ligate RNA to DNA in close proximity, followed by affinity purification of ligated complexes, library preparation, and sequencing.

(B) Schematic of methods used to identify RNA interactors of a given RBP. GoldCLIP and uvCLAP use affinity handles to purify crosslinked protein–RNA complexes, followed by adaptor ligation and direct elution of bound RNAs for sequencing. eCLIP, iCLIP, and irCLIP use protein-specific antibodies, SDS-PAGE, membrane transfer, and membrane excision to isolate RNAs for sequencing.

(C) Outline of methods to identify lncRNA-protein interactions. Left: methods to detect protein interactors of a candidate lncRNA. RAP-MS, ChIRP-MS, and CHART-MS chemically crosslink RNA to protein, use biotinylated probes to capture target lncRNA-protein complexes, and digest with RNase to release bound proteins for mass spectrometry; RaPID targets the BirA* biotin ligase to a stem-loop modified RNA to biotinylate protein interactors in live cells. Right: methods to identify new RBPs that bind to RNA in cells. RBDmap (left) and RBR-ID (right) use UV to crosslink RNA to protein in live cells in the presence of absence of 4SU. RBDmap lyses cells and employs two rounds of oligo(dT) capture to enrich for polyadenylated transcripts (left). RBR-ID isolates cell nuclei to identify all nuclear RNA-binding proteins. Both methods include RNAse digestion and protease digestion steps to generate short crosslinked peptides that can be analyzed by MS (bottom).

These methods have yielded robust results in select contexts, for example in the characterization of early spreading of XIST on the X chromosome [71, 163] and the identification of chromatin binding sites for MALAT1 and NEAT1 [151], but they suffer from high levels of background signal and contamination from crosslinked interactions [161], and alternative strategies might be required in cases where the biological material is limiting or the lncRNA of interest is expressed at low levels.

More recently, RNA-DamID was developed to map lncRNA-occupied sites in vivo [164]. This method is based on the principle of targeted DamID, in which the bacterial adenine methylase Dam is expressed in cells as a fusion to a protein of interest, and methylates adenines in the vicinity of its genomic targets [165]. In RNA-DamID, a lncRNA of interest fused to MS2 stem loops and the Dam enzyme fused to MS2 coat protein (MCP) are co-expressed in live cells. The stem-loop–MCP interaction recruits Dam to the lncRNA, and, therefore, to its binding sites on chromatin, which are methylated and subsequently identified by selective digestion with a methylation-sensitive restriction enzyme followed by sequencing (Figure 2A). RNA-DamID was applied to map tissue-specific binding of roX RNAs in Drosophila embryos with increased accuracy and sensitivity compared to previous methods, by identifying fewer false-positive peaks of roX binding and reducing the amount of required input material 100-fold [164]. The drawback of this method is the demand for genetic manipulation of cells and tagging the lncRNA of interest with MS2 stem-loops, preventing its usage in certain cases, such as patient-derived samples.

The techniques above are designed to map the localization of only a single lncRNA. It has now become possible to map the distribution of all chromatin-associated RNAs in an unbiased manner with the development of MARGI, CHAR-seq, GRID-seq, and more recently SPRITE [166–169]. MARGI, CHAR-seq, and GRID-seq use chemical crosslinking followed by a biotinylated oligonucleotide bridge to ligate DNA and RNA in close proximity, forming an DNA–RNA chimera that is captured and sequenced after reverse transcription (Figure 2A). SPRITE does not use proximity ligation but instead employs a split-and-pool strategy to sequentially label interacting DNA and RNA with barcoded tags, which are identified after sequencing by matching reads that contain the same combination of barcodes [168]. These methods have thus far uncovered local and long-distance DNA–RNA contacts in both human and Drosophila, and could shed light on the potential roles of lncRNAs in orchestrating 3D genome architecture, but, as with any unbiased approach, suffer from the limitations of being skewed toward detecting the most abundant species.

Defining protein–RNA interactions

Identifying the proteins that interact with a given lncRNA is critical for understanding its function. Protein–RNA interactions can be interrogated in vivo and in vitro by various binding assays, including RNA immunoprecipitation (RIP) with or without crosslinking, electro-mobility shift assays (EMSA), filter-binding assays, and fluorescence anisotropy measurements. These however, can only test candidate interactions where both the protein and the RNA are known. Several unbiased methods have been developed to discover new protein–RNA interactions and we divide them into those that identify all RNAs bound to a protein of choice (“RNA discovery”, Figure 2B) or choose a lncRNA and identify all proteins that bind to it (“protein discovery”, Figure 2C).

RNA discovery

Historically, RNA discovery methods were developed first. Immunoprecipitation in RNA-preserving conditions was coupled first to custom-designed microarrays interrogating large sets of RNAs simultaneously [9, 41, 170–172] and then with unbiased deep sequencing (RIP-seq, [173]). However, native RNA immunoprecipitation-based methods, lacking stringent washing steps, are known to suffer from exceedingly high noise levels [174], especially when combined with PCR-mediated amplification and sensitive deep sequencing. More recent alternatives are all based on variants of CLIP and its high-throughput version, HITS-CLIP (also known as CLIP-seq), which were originally developed to study factors regulating alternative splicing in the brain [175, 176]. CLIP entails UV-mediated crosslinking of RNA to its interacting proteins in vivo followed by immunoprecipitation of a protein of interest, separation on an SDS-PAGE gel and transfer to nitrocellulose membrane. Once the presence of crosslinked protein–RNA complexes on the membrane is confirmed at a position compatible with the expected molecular weight of the protein of interest, they can be excised and the eluted RNA can be processed for deep sequencing (Figure 2B).

Because of the covalent nature of protein–RNA crosslinking, some of the biochemical steps in CLIP are performed in extremely stringent conditions (i.e. SDS denaturation during SDS-PAGE) but the initial forms of this method were still troubled by high levels of noise, as well as poor yields [177]. Some of these issues were addressed in subsequent improvements. iCLIP introduced a circular ligation step after reverse transcription (RT) during library preparation, which allowed recovery of the truncated cDNAs resulting from early termination of RT at the crosslink site, increasing yields and providing single-base resolution information regarding the position of the presumptive protein–RNA interaction [178]. eCLIP further refined the iCLIP protocol and included a strategy to sequence size-matched input material in parallel [179], which is in our opinion a necessary step to calculate enrichment of true RNA interactors.

Other variants of CLIP have been developed to address the issue of low yields at the biochemical level, by replacing the highly inefficient and time-consuming membrane transfer and elution steps with in-solution purification strategies. GoldCLIP and uvCLAP are both based on epitope tags that allow for purifications in extremely stringent conditions [180, 181]. The GoldCLIP method involves fusion of the protein of interest to the bacterial HaloTag, which covalently attaches to chloroalkane-conjugated beads during purifications, whereas uvCLAP introduced an affinity tag that allows for in vivo biotinylation of the protein of choice followed by streptavidin-mediated purification of the crosslinked complexes under denaturing conditions. uvCLAP was utilized to study the nuclear RNA helicase DHXF9, revealing that it binds to RNAs containing inverted Alu repeats, proving the method’s practical advantages [182]. However, both GoldCLIP and uvCLAP rely on molecular engineering of the protein of interest to introduce the affinity tag, and therefore might not be suitable to investigate protein-RNA interactions in all contexts.

Finally, proximity labeling approaches [183] are now becoming available for RNA discovery. These are experimental strategies conceptually different from those described above, which all rely on some type of affinity purification of a protein–RNA complex. Instead, proximity labeling converts the physical proximity of potential RNA interactors to a protein of choice in vivo into a chemical labeling that can then be used to recover and identify the RNAs. The first of these approaches was in vivo proximity labeling (IPL), which utilized photoactivatable biotin to tag RNAs in the proximity of a protein of choice [184]. More recently, proximity labeling techniques developed to study protein–protein interactions have also been adapted for RNA discovery. APEX, which uses an engineered ascorbate peroxidase 2 to biotinylate targets [185] has now been used to biotinylate proximal RNAs in vivo [186, 187]. Alternatively, BioID uses a modified form of the BirA biotin protein ligase to biotinylate target molecules [188] and has recently been applied to RNA with the name of RaPID [189]. It is important to note that, like all proximity labeling strategies, these methods reveal proximity in vivo and not physical interaction and therefore caution should be used in inferring biochemical relationships between the protein of choice and the newly identified RNAs without additional validation steps.

Protein discovery

The first techniques for protein discovery (i.e., to identify proteins bound to a lncRNA) have been adaptations of the hybridization capture methods CHART, ChIRP, and RAP described above, whereby a protein elution step was added after the purification of crosslinked protein-RNA complexes followed by mass spectrometry (MS) for protein identification. Several methods based on these principles have been developed to date, including CHART-MS, ChIRP-MS, RAP-MS, and iDRIP [120, 121, 151, 190] (Figure 2C). ChIRP-MS, RAP-MS, and iDRIP were used to identify interactors of the Xist lncRNA in mouse embryonic stem cells, including the transcriptional repressor SPEN, which was functionally validated. RAP-MS was also applied to study the role of the lncRNA Sammson in promoting melanoma cell survival by stabilizing mitochondrial protein p32 [26], and the orchestration of a protein–RNA complex essential for genome stability by the lncRNA Norad [49].

Proximity-based methods are also available for protein discovery. For example, RaPID is an adaptation of BioID that uses the bacterial biotin ligase mutant BirA* to biotinylate the protein interactors of a given RNA motif in live cells [189] (Figure 2C). The RNA of interest is fused to bacteriophage BoxB stem loops and coexpressed with BirA* fused to the BoxB-binding λN peptide. The BirA*–λN fusion binds strongly to the BoxB stem loops and biotinylates proteins associated with the target RNA motif in vivo, which are purified and analyzed by MS.

Identification of new RNA-binding proteins and domains

In addition to candidate-based studies of proteins known to bind lncRNAs, several methods have been developed to identify new RNA-binding proteins in an empirical manner and, more recently, map their RNA-binding domains. Pioneered by the Hentze group, these approaches initially relied on protein–RNA crosslinking followed by affinity purification of polyadenylated transcripts and identification of crosslinked proteins by MS [142, 191–193]. A refinement of this method, termed RBDmap, allowed for the simultaneous identification of RNA-binding proteins and the mapping of the interaction site at peptide resolution [141]. We developed RBR-ID, a similar method, but based on a slightly different principle, which allowed us to identify a larger scope of RNA-binding proteins, including those that bind non-polyadenylated transcripts, and to map their RNA-binding regions in vivo [104]. Several other methods have been developed along these same lines, including CARIC and RICK, which combine metabolic labeling of RNAs and biotin conjugation for purification of crosslinked RNA and protein [194, 195], and XRNAX, which captures crosslinked RNA-protein complexes in the insoluble TRIZOL interphase during purification for downstream sequencing or MS [196].

Collectively, these methods have led to the realization that in addition to the several hundred “canonical” RNA-binding proteins, characterized by possessing one or more well-characterized RNA-binding domains (e.g. RRM, KH, dsRBD; [197]), hundreds—perhaps thousands—more form protein–RNA interactions in vivo, whose functional significance remains unexplored. We propose that in many cases these interactions might be taking place with lncRNAs and that characterizing them will greatly contribute to our understanding of lncRNA function. Furthermore, “targeted” versions of RBR-ID and RBDmap have been developed that can map protein–RNA interactions at high resolution within a complex of choice (e.g. PRC2 [198]). These studies will be instrumental to design “separation-of-function” mutants to address the biochemical and biological roles of RNA interactions within these complexes and validate the proposed biological functions of lncRNAs (see above).

RNA–RNA interactions

While small noncoding RNAs—including micro RNAs, Piwi-interacting RNAs, and small nucleolar RNAs—are known to function largely via RNA–RNA base pairing [43], whether lncRNAs also participate in functional interactions with other RNAs is much less clear.

RAP-RNA was one of the first methods developed to identify the RNA interactors of specific lncRNAs. In RAP-RNA, cellular RNA is chemically crosslinked with one of three crosslinking agents to enrich for direct, both direct and indirect, or only indirect RNA interactions [199]. This is followed by capture of a target lncRNA using antisense biotinylated probes. While RAP-RNA revealed that the abundant MALAT1 lncRNA interacts with pre-mRNAs at active chromatin loci, it recovered virtually no direct RNA interactors, suggesting that MALAT1 does not base pair with its mRNA targets.

CLASH is another recent method that has enabled the high throughput sequencing of direct RNA–RNA interactions [200]. CLASH requires the expression of a tagged “bait” protein in cells, followed by UV crosslinking of RNAs to protein. The tagged protein is pulled down with its crosslinked RNAs, which are ligated to form chimeric molecules that are subsequently sequenced. CLASH has been applied to identify global miRNA interactions in human cells, revealing widespread non-canonical binding of microRNAs to mRNAs and ncRNAs. CLASH may hold promise for identifying lncRNA–RNA interactions, but is limited by the requirement of a universal protein bait. Two additional methods—PARIS and LIGR-seq—allowed for the mapping of RNA duplexes in living cells through a combination of psoralen crosslinking and proximity ligation followed by sequencing [201, 202]. The resulting data can be used to infer RNA secondary structure as well as functional interactions. COMRADES, a similar method, uses tandem affinity purification to isolate RNA duplexes, and was developed alongside a new computational pipeline to predict and cluster RNA structures [203]. Because these methods identify global RNA interactions, it may be challenging to leverage them to study lncRNAs, many of which are lowly expressed.

Aiding biochemistry: genetic and computational tools to study lncRNAs

Although the many new methods discussed above have the potential to reveal a large number of new biochemical interactions of lncRNAs, often choosing which lncRNA to study is the most difficult question, and biochemistry alone might not provide the answer. The number of annotated lncRNAs in the human genome is in the order of nearly ten thousand [6], and in many cases the lncRNA itself has no discernible RNA-mediated function [204]. How, then, can the needle of a new functional lncRNA be found in the haystack of thousands of potentially non-functional annotations? New genetic and computational tools might provide the answer.

Genome-wide forward genetic screens

Forward genetic screens in mammalian cells have become feasible due to the power and versatility of CRISPR-based tools, which allow for both loss- and gain-of-function screening strategies in a pooled format [205] (Figure 3A) allowing the direct interrogation of thousands of lncRNAs in a relatively manageable experiment.

Figure 3. Functional characterization of lncRNAs.

(A) CRISPR screens have been used to identify functional lncRNAs. A guide RNA library is created against a set of lncRNA genes and transfected into cells, with the goal of targeting one gene in each cell. Different proteins can be targeted to the locus of interest, including Cas9 (knockout screens), a combination of NLS-dCas9-VP64 and MS2-p65-HSF1 (synergistic activation mediator, “SAM”; CRISPR activation screens), or dCas9-KRAB repressor (CRISPR interference screens). Following selection of successfully transfected cells, cells are allowed to proliferate, with the possible addition of a treatment such as a drug. After a period of growth, the composition of guide RNAs in the final pool of cells is compared to that in the starting population. Enriched or depleted guides indicate a phenotype for the targeted gene, depending on the direction of enrichment and the type of screen.

(B) Some lncRNAs exhibit synteny without conservation. In these cases, two lncRNAs that occupy the same genomic position in different species do not have significant sequence similarity, suggesting that these lncRNAs might act locally to regulate transcription of neighboring orthologous protein-coding genes.

(C) LncRNAs can be grouped functionally by their k-mer profile. In this example, activating lncRNA 1 and 2 have similar k-mer profiles, while repressing lncRNA 3 and 4 have similar k-mer profiles.

The first lncRNA loss-of-function screen utilized CRISPR interference [206] to identify lncRNAs required for survival in various cell lines [207]. Consistent with the marked cell type specificity of these noncoding transcripts, 89% of lncRNAs with a growth phenotype were specific to one cell type, compared to 55% of protein-coding mRNAs, and no single lncRNA was required for survival in all cell types. Critical pathways such as translation, DNA replication, and post-transcriptional regulation were perturbed by lncRNA repression. Functional lncRNAs identified by this screen have a set of distinct characteristics: expression level, distance from an enhancer, and number of exons were predictive of a growth phenotype following their repression. A gain-of-function screen identified lncRNAs that conferred resistance to BRAF inhibitors in melanoma cells [126]. One hit from this screen was the lncRNA EMICERI, which is responsible for the activation of the nearby protein-coding gene Mob3b, part of the vemurafenib resistance-associated Hippo signaling pathway. Most lncRNA hits that mediated BRAF resistance acted by affecting the expression of nearby genes, adding to the number of examples of lncRNAs that regulate targets locally.

Genetic screens have also been used to identify lncRNAs with positive or negative effects on growth in HuH7.5 liver cancer cells [208], lncRNAs that affect a specific pathway [209], and antisense lncRNAs that regulate cell viability following Ara-C chemotherapy in acute myeloid leukemia (AML) [210]. The AML screen found one coding/noncoding pair, GAS6/GAS6-AS2, that also had correlated expression in a large database of cancer cells treated with Ara-C. Overexpression of the lncRNA GAS6-AS2 caused the upregulation of its cognate protein-coding gene GAS6 locally but also induced expression of the GAS6 receptor gene, AXL, distally, suggesting that this lncRNA might act via multiple biochemical mechanisms.

Computational approaches to infer lncRNA function

Classic tools used to infer the functional importance of protein sequence, namely its evolutionary conservation, are less powerful in the context of lncRNAs, which are not under pressure to maintain a coding frame. However, rather than maintaining primary sequence through evolution, lncRNAs may conserve their function through positional homology, k-mer content, or secondary structure.

Many lncRNAs have synteny without sequence conservation, perhaps indicating a regulatory role that does not depend on sequence. For these genes, sequence conservation exists in protein-coding genes and genomic sequences flanking the lncRNA locus, but not in the lncRNA itself (Figure 3B) [211].

Also suggested is a conservation of short sequence motifs, or k-mers, as the frequency of certain k-mers may impact the ability of a protein to bind to the transcript [212]. The k-mer profiles of lncRNAs from different functional groups, such as those that repressive or activate transcription locally, tend to cluster together (Figure 3C). The localization and binding ability of proteins can be predicted from k-mer profiles. The dynamics of lncRNA biogenesis can also be used to infer functional classes of lncRNAs. Measuring transcription, splicing, degradation, localization, and translation dynamics of protein-coding and noncoding RNAs recovered known characteristics of lncRNAs, including slower splicing, synthesis, and processing rates, paired with higher degradation rates [213]. The function of lncRNAs were assigned by grouping lncRNAs with protein-coding genes of known function that had similar characteristics. This produced protein-coding/lncRNA groups with a wide range of metabolic properties and distinct likely functions.

Another common strategy for identifying lncRNAs and their function is correlating the expression of lncRNAs and protein-coding genes. Increasing amounts of sequencing data from various tissues and conditions allow for the detection of highly variable genes and the identification of frequently coexpressed genes. The function of a lncRNA can be inferred from the identity of protein-coding genes that have a similar transcriptional profile [10]. In non-model organisms, where sophisticated genetic tools are not always readily available, the approach of using large sequencing datasets to identify correlated lncRNAs and protein-coding genes has promise to prioritize the study of functional lncRNAs.

Finally, comparative genomics and evolutionary analyses of lncRNA sequence conservation have provided information on potential lncRNA function. Co-expression networks of lncRNAs and protein-coding genes that are expressed in multiple organisms helped identify lncRNAs involved in spermatogenesis, synaptic transmission, and muscle functions in the heart [10]. Although lncRNAs evolve more rapidly than protein-coding genes, lncRNA promoters contain more transcription factor binding sites than random intergenic regions do, and are particularly enriched for motifs recognized by homeobox transcription factors. The age of a lncRNA affects the likelihood of its active regulation, as the Polycomb protein SUZ12 and the pluripotency transcription factor OCT4 associate more with older, more conserved lncRNAs [10].

Concluding remarks/future outlook

We have witnessed an acceleration of new technologies and strategies to discover and characterize functional lncRNAs, several of which are reviewed above. Looking ahead, continued improvements to biochemical and genomic technologies will allow for the functional interrogation of lncRNAs in specific tissues and cell-types, as well as the identification of novel lncRNAs with important biological roles.

Given their very specific and restricted expression patterns, we believe that great insight into lncRNAs function will be offered by single-cell sequencing technologies [31]. Indeed it has already been shown that at least in the case of some lncRNAs their presumptive lower expression level in bulk samples was a consequence of the smaller population of cells in heterogenous mixtures that expressed them. A collection of 81 single-cell transcriptomes revealed that lncRNA are activated during reprogramming and suppress lineage-specific genes on a single-cell level [214]. Another study of 226 cells from radial sections of the mouse neocortex identified cell type-specific lncRNAs [214]. Single-cell sequencing has revealed different lncRNA profiles among fibroblast cell types [215] and between neurons and glia in Drosophila [216]. Although these studies are informative, higher-throughput studies may provide more information about the specificity and regulatory potential of lncRNA in the context of a whole tissue or organism.

Evolutionary studies of lncRNAs have been aided by new annotations of lncRNAs in a variety of model and non-model organisms, facilitated by new and improved genome assemblies. For example, new lncRNA annotations in Xenopus tropicalis and the analysis of their expression patterns suggested functional roles in tissue identity during development [217]. Annotations of lncRNAs have become available in multiple social insect species [15, 16], where alternative developmental trajectories give rise to individuals with identical genomes but vastly different morphological, physiological, and behavioral phenotypes. We analyzed lncRNA expression in Harpegnathos saltator, an ant that maintains reproductive plasticity in workers during a unique caste transition [218]. Major transcriptional changes occur in the brain during this event, including differential regulation of lncRNAs [16, 219]. As CRISPR-mediated editing of the germline has been achieved in Harpegnathos ants [220], these newly discovered caste-specific transcripts are prime targets for the functional investigation of lncRNAs potentially involved in brain function and behavior.

Another emerging frontier in lncRNA research is the analysis of their post-transcriptional chemical modification [221]. Pioneering studies on XIST showed that its modification with m⁶A is required for the silencing of several X-linked genes [222]. DCI, a protein required for XIST-mediated transcriptional silencing, binds preferentially at m⁶A residues. As folding of lncRNAs is thought to be key to their function, it will be of great interest to determine if chemical modification that affect their three-dimensional structure contribute to their biological and biochemical versatility.

Given the many known and suspected roles of lncRNAs in various cellular processes, it should not come as a surprise that dysregulation of lncRNAs can lead to disease or aberrant phenotypes. LncRNA are increasingly being considered as therapeutic targets [223, 224]. As mentioned above, genetic screens have detected many lncRNAs involved in cancer progression [126, 209, 210]. Additionally, organisms that transmit or cause disease can affect how the immune system recognizes threats through regulation of lncRNAs in their own transcriptome or in host cells. The genomes of mosquitoes and of the malaria parasites harbor lncRNAs [225–228], and the parasite Toxoplasma gondii affects the lncRNA profile of its host [229]. The annotation of lncRNAs in more organisms will lead to better understanding of their function and will provide new avenues to study lncRNA-related diseases and their treatment.